pandas-dev/pandas · error · AssertionError

{obj} are different {message}

Error message

{obj} are different

{message}

What it means

raise_assert_detail (asserters.py:694-735) is the central failure-reporting function for pandas.testing assertions. It assembles a structured message — object name, what differs, [index], [left], [right], optional [diff] and first-diff — and raises AssertionError. It is the body you see when assert_index_equal, assert_series_equal, assert_frame_equal, assert_numpy_array_equal, or assert_extension_array_equal find an actual inequality.

Source

Thrown at pandas/_testing/asserters.py:735

    elif isinstance(left, (CategoricalDtype, StringDtype, NumpyEADtype)):
        left = repr(left)

    if isinstance(right, np.ndarray):
        right = pprint_thing(right)
    elif isinstance(right, (CategoricalDtype, StringDtype, NumpyEADtype)):
        right = repr(right)

    msg += f"""
[left]:  {left}
[right]: {right}"""

    if diff is not None:
        msg += f"\n[diff]: {diff}"

    if first_diff is not None:
        msg += f"\n{first_diff}"

    raise AssertionError(msg)


def assert_numpy_array_equal(
    left: Any,
    right: Any,
    strict_nan: bool = False,
    check_dtype: bool | Literal["equiv"] = True,
    err_msg: str | None = None,
    check_same: Literal["copy", "same"] | None = None,
    obj: str = "numpy array",
    index_values: Index | np.ndarray | None = None,
    *,
    class_obj: str | None = None,
) -> None:
    """
    Check that 'np.ndarray' is equivalent.

    Parameters

View on GitHub (pinned to 71959b8cb9)

Solutions

1. Read the [left]/[right]/[diff] blocks in the message to locate the first divergence.
2. If the difference is floating-point noise, pass check_dtype=False and use rtol/atol (or check_exact=False).
3. If index/names differ, set check_names=False or check_index=False only if intentionally skipping that check.
4. Fix the data or the expected fixture so the objects truly match.

Example fix

# before — fails on float dtype/values
assert_series_equal(pd.Series([1.0]), pd.Series([1.000001]))

# after
assert_series_equal(pd.Series([1.0]), pd.Series([1.000001]), check_exact=False, rtol=1e-4)

# Decide on exactness/tolerance before asserting
import numpy as np
if left.dtype.kind in 'iu' and right.dtype.kind in 'iu':
    check_exact = True
else:
    check_exact = False  # use rtol/atol for floats

try:
    assert_series_equal(left, right, check_exact=False, rtol=1e-5, atol=1e-8)
except AssertionError as e:
    # log the structured diff for diagnosis
    raise

When it happens

Trigger: Two pandas objects that are genuinely unequal: differing values, differing shapes/lengths, differing dtype when check_dtype=True, differing names when check_names=True, or differing categorical categories. raise_assert_detail is invoked from inside the assert_*_equal functions after a comparison fails, so hitting it means a real test failure.

Common situations: Float comparisons without rtol/atol; dtype drift (int32 vs int64); timezone-naive vs timezone-aware datetimes; index alignment differences; categorical category ordering; the expected fixture was updated but not the assertion.

Related errors

• {left_base!r} is not {right_base!r}
• {left_base!r} is {right_base!r}
• [datetimelike_compat=True] {left._values} is not equal to {r
• {cls_name} Expected type {cls}, found {type(left)} instead
• {cls_name} Expected type {cls}, found {type(right)} instead

AI-assisted analysis of pandas-dev/pandas@71959b8cb9 (2026-08-07). Data as JSON: /api/errors/34b49ab37ad8de7c. Report an issue: GitHub.